Generalizable Protein Dynamics in Serine-Threonine Kinases: Physics is the key

Abstract

Kinases play crucial roles in signaling pathways across oncology, inflammation, and neurodegenerative diseases. Historically, their conformational states have been defined by the DFG motif: DFGin and DFGout. However, this binary paradigm overlooks the broader conformational heterogeneity of apo kinases, which encompasses multiple metastable states within an expanded DFG-Phe ensemble. We introduced a novel nomenclature that integrates the dynamics of the DFG-Phe, activation loop, and C-helix, highlighting how these regions respond to various ensemble perturbations (e.g., mutations, ligand binding, and protein-protein interactions). A major bottleneck in studying kinases lies in sampling their wide-ranging conformations because static snapshots often remain trapped in specific free energy minima, limiting traditional molecular dynamics simulations from exploring multiple functionally relevant states starting from a single structure. To accelerate conformational sampling, we introduce a computational framework that integrates AlphaFold, machine learning, physics- based simulations, and Markov state modeling. Rather than focusing on single-structure snapshots or structural hypotheses generated by protein structure prediction models, our framework captures shifts in conformational populations under varying perturbations, shedding light on both the thermodynamics and kinetics of the transitions. We show generalizability of our ensemble definitions and protocol across members of serine-threonine kinases and tyrosine kinases. A key innovation lies in our machine learning algorithms, which capture slowly varying structural features to uncover hidden states and generate latent representations of conformational motions across different kinase domains. The physics-refined structural ensemble sampled from the latent layers is then used to launch new simulations that more comprehensively explore the full conformational landscape than traditional molecular simulation approaches. By capturing how these conformational shifts influence downstream protein-protein interactions, conformational allostery, and cryptic pocket formation, our accelerated simulation framework provides deeper insights into generalized molecular recognition mechanism in kinases and how conformational heterogeneity is influenced by ensemble perturbations. This framework can be extended to investigate broader protein families--such as G protein-coupled receptors (GPCRs) and tumor necrosis factors (TNFs)--where functional outcomes are dictated by conformational heterogeneity.